After all processes migrate to a single one one must move all metadata into the overall process'es heap. There is another metadata consumer that resides in MDB plugin code.
The consumer must become another "workernode client"(MCS specific consensus proto implementation node) Here are the related functionality that the consumer do:
- translates from MDB's query tree representation into MCS analog.
- provides metadata(EM, partitioning) visibility via SQL.
The first activity needs syscat contents: tables, columns so the consumer type needs only syscat state that is consistent with the cluster. The state can be even eventual.
The consumer can query the new overall columnstore process for additional metadata, e.g. extentmap and partitioning details.